HyRe personalizes reward models at test time by reweighting an ensemble of heads trained on aggregate preferences, using few target examples to outperform uniform averaging and prior methods on RewardBench and 32 tasks.
Diversify and disambiguate: Learning from underspecified data
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2024 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
Test-Time Alignment via Hypothesis Reweighting
HyRe personalizes reward models at test time by reweighting an ensemble of heads trained on aggregate preferences, using few target examples to outperform uniform averaging and prior methods on RewardBench and 32 tasks.